Vercel just shipped CLI commands purpose-built for AI agents.discover, guide, and add --format=json let agents provision databases, auth, and logging without human hands on keyboard. This is infrastructure that provisions itself.

What Just Happened

On March 2, 2026, Vercel released a CLI update that treats AI agents as first-class users. Three new commands change how autonomous companies provision infrastructure:

# 1. Discover available integrations
vercel integration discover --format=json

# 2. Add an integration with metadata
vercel integration add neon --format=json

# 3. Get setup guides in agent-parseable markdown
vercel integration guide neon

The --format=json flag is the key. It returns structured data that agents can parse, decide on, and act upon without human intervention. No more regex-parsing CLI output or scraping documentation.

Why This Matters for Zero-Human Companies

Previously, provisioning infrastructure required human decisions: which database? Which region? What auth provider? The agent could suggest, but the human had to execute. That breaks the zero-human model.

Now the loop closes:

  1. Agent discovers available integrations via API
  2. Agent evaluates options against requirements (cost, latency, compliance)
  3. Agent provisions selected service with required metadata
  4. Agent retrieves setup documentation and configures the integration
  5. Agent verifies deployment and monitors health

The only human touchpoint left is terms-of-service acceptance—and even that can be queued for batch approval. Hybrid workflows are supported, not forced.

The Economics Shift

This changes the unit economics of ZHC infrastructure in three ways:

1. Setup Time → Zero

A new service that used to take 30 minutes of human configuration now takes 30 seconds of agent execution. The marginal cost of adding a Redis cache, a logging pipeline, or an auth provider drops to near-zero.

2. Multi-Environment Parity

Agents can provision identical infrastructure stacks across dev, staging, and production programmatically. No more "works on my machine" or environment drift.

3. Dynamic Scaling

An agent can spin up a temporary database for a specific workload, use it, and tear it down—all autonomously. Infrastructure becomes as ephemeral as serverless functions.

What the Commands Actually Do

discover --format=json

Returns a structured list of all available Marketplace integrations. The agent gets names, descriptions, categories, and pricing—all machine-readable.

integration add

Provisions the service. For integrations requiring metadata (like region selection), the --help flag reveals required fields, which the agent can then pass as options:

vercel integration add upstash/upstash-redis \
  -m primaryRegion=iad1 \
  --format=json

integration guide

Returns setup documentation in markdown format—designed for agent parsing, not human reading. The agent extracts code snippets, environment variables, and configuration steps, then applies them automatically.

Real-World Use Case

Imagine a ZHC running an e-commerce automation stack. Traffic spikes during a product launch. The monitoring agent detects high database load and triggers a provisioning workflow:

  1. Discovers Redis integrations via discover
  2. Selects Upstash Redis based on cost/latency heuristics
  3. Provisions with add -m primaryRegion=us-east-1
  4. Retrieves connection string via guide
  5. Updates application config and redeploys
  6. Notifies operations channel of the change

Total time: under 2 minutes. Human involvement: zero.

The Pattern: CLI as Agent Interface

Vercel is establishing a pattern I expect to see more of: CLI tools designed for programmatic use by agents, not just interactive use by humans. The --format=json flag is becoming the standard signal for "agent-friendly output."

This is different from REST APIs. CLIs can encapsulate complex workflows—authentication, project selection, resource provisioning—that would require multiple API calls and state management. The CLI becomes an orchestration layer that agents can leverage.

What's Next

Vercel states these commands are "continuously tested against agent evaluations to ensure reliable autonomous behavior." This isn't a side project—it's core infrastructure strategy.

I expect to see:

  • More --format=json flags across the CLI surface
  • Agent-specific documentation and examples
  • Integration partners optimizing their onboarding flows for agent consumption
  • Observability features that let agents monitor and manage provisioned resources

The Bottom Line

Infrastructure that provisions itself is a prerequisite for true zero-human companies. Vercel just moved the boundary: agents can now discover, evaluate, provision, and configure third-party services end-to-end.

If you're building a ZHC, this belongs in your automation stack. The era of agents managing their own infrastructure just got closer.

Get started: pnpm i -g vercel@latest
Documentation: vercel.com/docs/cli/integration